loupvaillant/monocypher-handshake

Secure channels with Monocypher

monocypher-handshake's Issues

Check the design with ProVerif (or was it CryptoVerif?)

Protocol design turns out to be more subtle and error prone than I had anticipated. I cannot trust myself with this, I need external review. In the mean time, ProVerif (or CryptoVerif) looks like it could help.

Unsafe use of Poly1305

I have noticed that on empty inputs, the result of Poly1305 is equal to the last 16 bytes of the key. Meaning, using Poly1305 even once can reveal up to 128-bits of the authentication key.

When the input is not empty (in this case, 2 blocks), the leak is not so obvious, but there's still a relation between the first part of the key, and the second part of the key. This relation can probably be used to reduce the search space, and reduce the effective entropy of the key so something much below 256-bits.

Thus, it is probably a very bad idea to derive further key material with a key that has been used with Poly1305, so instead of

T2 = Poly1305(K2, Es || Er)
T3 = Poly1305(K3, Es || Er || Chacha20(K2, Ls))

We probably want something like:

T2 = Poly1305(HChacha20(K2, 0), Es || Er)
T3 = Poly1305(HChacha20(K3, 0), Es || Er || Chacha20(K2, Ls))

Also, Ks should probably be HChacha20(K3, 1) instead of just K3, to provide a clean key with no used nonce (K3 uses nonce 0 to derive an authentication key, which the user might reuse by accident).

Alternatively, we could derive a whole Chacha20 block with K3, use the first half for T3, and the second half for Ks. This would save one Chacha20 round. If we do that, though, T2 should probably be handled the same way, for uniformity's sake. (Another benefit would be that this is the standard way to derive Poly1305 keys authentication keys with Chacha20.)

Interoperability with Noise

To what extent is Monocypher-Handshake compatible with third-party implementations of Noise proper, has this been tested, and against which implementations, for what handshake patterns, etc? Thanks!